Here are my designs for the ReSpeaker Lite (short square) and ReSpeaker XVF3800 (tall round). Both have a 2" speaker and 2.5" passive radiator from parts-express.
I’ve noticed that the volume output of the XVF3800 is considerably lower than the Lite.
I love the design with the fabric! Been trying to design something simular myself but my one year old is apparantly more important Would you mind sharing those lovely 3D files?
Uh, I see you use up-firing speaker (at least for XVF3800 that is). The main goal is to isolate board as much as we can from the influence of speaker. Don’t know if it won’t be prone to detection problems…
@formatBCE Love the YAML integration you created. I had to build it on a docker container on my Host as there are some incompatibilities in HA at the moment with pydantic. Was just wondering, if it is possible to integrate OpenWakeWord instate of MicroWakeWord and may be an option for a continous conversation when using it with ChatGPT
Hi, Thanks a lot for the integration. I finally got it working.
Did update the USB firmware to 2.0.5. Thought it was newer and would work better than the I2S. But I was wrong, today I updated the I2S and everything finally works!
I had an old Google Home Mini laying around, stripped it for parts.
As I read the speaker is a 5W speaker it’d be perfect for the reSpeaker!
I stripped the connector and soldered a JST one on. Perfect fit!
Now I need to design a 3d printed case with clear PETG filament for the LEDs
We’re friends with FutureProofHomes.
Sat1 has a bit different approach, although the hardware is similar. Sat1 is using open source version, XU-316 (similar to Respeaker Lite and Voice PE). Which is also used in XVF3800, but with proprietary solutions from XMOS.
We’re yet to see if it’s better. But I know that Sat1 still uses two of four mics. This one is using all four. The main downside now is in DFU firmware - Seeed devs are working on harnessing XMOS SDK for best results.
Sorry i may have missed your message!
IDK i have no incompatibilities…
In theory you could use Openwakeword. But that would require constant streaming to HA. Also it would require major rework in YAML, since things like “stop” word wouldn’t be possible.
Continuous conversation, on the other hand, should work. In this manner current device is pretty similar to Voice PE and Respeaker Lite.
Has anyone done a direct comparison between the ReSpeaker Lite and the XVF3800? It would be helpful to hear samples recorded in a noisy room at different distances using both devices.
I’m currently exploring the possibility of integrating a noise suppressor into the STT server, but I’m not entirely satisfied with the data obtained from RS Lite. That’s why I’m interested in how effectively a “virtual directional microphone” is implemented on an array of 4 mics.
It doesn’t make sense so far. Seeed devs are still working out on the best approach to use XMOS proprietary technology to get the most suitable refinements to mic stream. So if you curious - you can do it. But it will change.
To do this, add following to your configuration.yaml and reboot HA:
assist_pipeline:
debug_recording_dir: /share/assist_pipeline # or your dir
Then you will have recordings from interaction in that folder.
Not sure how to check channel that goes to wake word (they’re different), but you could switch channels in YAML, for example.
I don’t have the device (as you said, it’s not time to do that yet). That’s why I’m asking if anyone has done a direct comparison between these two boards? I know how to get audio, though it’s useful info if someone wants to test.